ollama - 💡(How to fix) Fix Still crashing on Ollama 0.17.1 with Intel Arc 140V (iGPU, Vulkan) [2 comments, 2 participants]

Exception 0xc0000005 in ggml_backend_sched_graph_compute_async rax=0x0 (null pointer dereference)

Fix Plan

To resolve the null pointer dereference issue in ggml_backend_sched_graph_compute_async, we need to ensure that all pointers are properly initialized before use.

• Update the ggml library to the latest version to ensure any known issues are fixed.
• Verify that the Vulkan pipeline is correctly set up and that the coopmat disable flags are properly applied.
• Modify the code to check for null pointers before calling ggml_backend_sched_graph_compute_async.

Example code snippet to check for null pointers:

if (graph != NULL && sched != NULL) {
    ggml_backend_sched_graph_compute_async(graph, sched);
} else {
    // Handle the error, e.g., log and exit
    printf("Error: Null pointer encountered\n");
    exit(1);
}

Verification

To verify the fix, run the application with the updated code and check for the null pointer dereference error. If the error persists, further debugging may be required to identify the root cause.

Extra Tips

• Ensure that the Vulkan SDK and drivers are up-to-date.
• Consider adding additional error checking and logging to help diagnose any future issues.
• If using Docker, verify that the container has the necessary dependencies and permissions to access the GPU.